Towards a Safer and More Robust Internet Paul Vixie Internet Software Consortium August 2003

Success Disasters An idea can work well in a laboratory –everything works as it should And in testing –Small, controlled population –Limited patterns of use But fail in the broad public marketplace –6 is not 60 is not –Some malicious users or observers

Example Success Disasters Internet HOSTS.TXT file –Listed every internet-connected host of the day –Impossible to update, distribute, search at scale –Thus: DNS Classful Internet addressing –100 A, B, C networks –Impossible to create/maintain routing table –Thus: CIDR

More Examples 4 billion hosts allowed in IPv4 –Not enough for each person on Terra –Definitely not enough for each cell phone –Thus: IPv6 Networks operated by right-of-way owners –ISPs operate networks with leased assets –ISPs are more plentiful than Telcos –Thus: Internet Exchange Points

Even More Examples DNS protocol only allows 13 root servers –Unforeseen result of early IPv4 design choices –Host population demands vastly better coverage –Thus: anycast Dot Com bubble made many millionaires –Investors bought stock for resale, not dividends –Not every 20-something can be a millionaire! –Thus: crash and recession

Current Examples ISP operators must be experts in their field –Very little off the shelf technology or training –Vendor defaults are completely inappropriate –Thus: no admission control on customer traffic Host operators must be expert technologists –The Internet is full of dangerous places/people –PC and DSL market allows universal access –Thus: wide scale vulnerability and abuse

Problem Size and Scope The Internet doubled in size every 16 months from 1990 to 1999 Its not even theoretically possible to map it –Or list vulnerable hosts (or fix them) –Or keep track of bad actors (or arrest them) The Internet has become a new ecosystem –Like biology: can be studied but not understood

Necessary Next Steps (1) Make infrastructure stronger, more robust Increase technology training by 1000x –Starting with students still in school –Including vendor engineers –Including ISP operators –Including home users Note: any country can still compete

Necessary Next Steps (2) Make mirror copies of important servers –F.root-servers.org now in Palo Alto, Madrid, Hong Kong, Los Angeles, New York City, Seoul, Auckland, Rome, Sao Paulo; coming soon to a few dozen more cities Add security to core Internet protocols –DNS was never secure, but it can be made so –ISC and others are very active in this area

Necessary Next Steps (3) Increase interconnection between ISPs –Stop depending on few/large intermediaries –Every metro needs its own IP exchange –Regulated carriers must be required to join Teach moral philosophy (ethics) in schools –Most Internet attacks are by children/teenagers –Damage done feels remote and theoretical –Many literally do not know right from wrong

Probable Next Steps All of this will cost more money –Increase size and training level at ISPs –Increase training available to home users –Increase interconnectivity between ISPs Internet usage fees are artificially low –Cost models are still immature –Many services are sold below current costs

Conclusions Biggest Internet engineering problem, ever: –Scaling Cyberspace and meatspace are now merging –Problems in either are really problems in both Weve got to take all of this more seriously –Internet vulnerability is a problem for society Its not a sandbox or a laboratory any more –Safety is more important than money, or fun